Gavin Baker: Anthropic's $2T IPO, Zuck's AI Manifesto, Nvidia's $500B AI Bet, Grok's Comeback

The episode opens with a structural joke that turns out to be the episode's organizing principle: a thin bench and a fat agenda. Jason frames the show as "a short crew, but a long docket," with only David Sacks and Gavin Baker alongside him, and then immediately tests whether the guest-of-honor framing still applies to Baker after SpaceX. The answer is a graceful deflection rather than a victory lap: SpaceX was "a magical moment," but "there's only one SpaceX," and the work continues. Baker borrows Bill Belichick's post-victory line about moving on to Cincinnati, which is exactly the right register for a show that treats capital allocation as a sport with a season.

The Rahm Emanuel interview detour is not filler. It sets up the political frame the hosts return to for the rest of the hour: the argument inside the Democratic coalition between institutional moderates and the democratic-socialist wing. Sacks reads Emanuel as a throwback to the Clinton-Obama party, and Jason's summary of the interview is that Emanuel went after democratic socialism as a road to nowhere, drawing on a résumé that includes the Obama and Clinton administrations, the Chicago mayoralty and an ambassadorship to Japan. The hosts' shared hope is explicitly partisan-adjacent rather than neutral: they want moderates to have a chance, and they say so.

What is notable here is the analytical move underneath the banter. The hosts treat the internal fight in one party as a preview of a fight they think is coming in the other, and they use the word "civil war" for it. That framing matters later, when the conversation turns to AI governance and the question of who gets to decide how a general-purpose technology is distributed. The same instinct that makes them prefer contested primaries makes them prefer contested model markets.

The lead story is a Financial Times report that Anthropic is targeting a two-trillion-dollar IPO, which would clear SpaceX's record-setting $1.75 trillion listing. Jason adds the corroborating detail that Polymarket prices an Anthropic IPO this year at roughly 80%, and that Brad Gerstner has been signalling the same direction. The reported timing is October, six to eight weeks out from the recording, with the show's own convention of taping Thursday and publishing Friday stated up front so listeners can date the claim.

The revenue side is where the numbers get strange. The episode cites an annualized run rate ending the year between $100 billion and $120 billion, a figure the hosts have tracked across several episodes, against a $2 trillion valuation. Jason does the multiple out loud: 16 to 20 times sales, which he notes is a fraction of what SpaceX, Palantir and other richly valued names went out at. The growth rate cited is 10x over the past year, and the claim attached to it is that Silicon Valley has never seen a ramp of this shape.

Two competitive footnotes are folded into the same breath. Polymarket gives Anthropic a 67% chance of holding the top model at year end, which concedes that the lead is no longer uncontested; Grok is described as catching up quickly and OpenAI as "not too shabby." The same segment carries a separate breaking item: Anthropic in talks to buy an AI startup called Descartes for $6 billion, software that lowers training and inference costs by making chips more efficient. That acquisition, if real, is a tell about where Anthropic thinks its binding constraint sits.

Jason then does the thing good interviewers do: he asks the guest to disclose his position before answering. Baker's answer is clean and load-bearing for everything he says afterwards. He is not currently a shareholder, he says he generally does not invest in Elon's competitors, and he calls that a deliberate policy rather than an accident. It is a useful reminder that the most quoted bulls on this show often have no direct economic exposure to the name they are defending.

Baker's contribution to the IPO discussion is mostly epistemics rather than valuation. He is "a little skeptical" of the two-trillion figure, and his reason is institutional: when every bank is jockeying to be lead left, the losers have an incentive to leak a number that makes the winner look bad. On that theory, the reported $2 trillion is not a forecast but a hurdle planted in public by someone who did not get the mandate.

Jason supplies the memory that makes the point concrete, reminding Baker that he had previously floated a $3 trillion to $4 trillion outcome, which Baker accepts as roughly where the stock might trade rather than where it would be priced. That distinction, pricing versus trading, is the hinge of the whole segment. Jason names the practice: "So they're sandbagging now." Baker agrees, then explains why a responsible issuer would do it anyway, pricing low enough to absorb the lockup rather than creating volatility that distracts employees at the worst possible moment.

The falsifiable version of the claim is stated plainly: if the deal prices at $2 trillion, it means the testing-the-waters and the roadshow went unusually well, because people genuinely want to own the asset. Baker also shuts down a possible misreading, confirming that no roadshow has happened yet. What makes the segment worth keeping is that it gives the listener a way to interpret the eventual number rather than just a prediction to score.

There is an unstated limitation worth naming. The whole analysis assumes a conventional book-building process with a visible lead-left bank and a normal roadshow, and it assumes the leak economy behaves the way it did in the 2010s. None of the participants offer evidence for the leak's provenance beyond pattern recognition, and the $2 trillion figure is never reconciled with the 16-to-20-times-sales math established minutes earlier. At a $120 billion run rate, $2 trillion implies roughly 16.7x; the multiple and the leak are consistent, but nobody says so out loud.

Sacks takes the growth question seriously instead of treating it as a slogan. If the company ends the year near $100 billion and has grown 10x year over year for three consecutive years, then simply continuing the trend produces a trillion dollars of ARR by the end of next year. He immediately converts that into a constraint problem rather than a demand problem: is the TAM big enough, is there enough compute, is there enough energy. His own answer is that demand is not the binding constraint and physical inputs are.

The evidence he offers for durable demand is the expansion of use cases rather than a forecast: agents are only now taking off, GrokBot's launch is cited as a marker, and every additional context an AI is dropped into adds token demand. That is a plausible mechanism, but it is also unfalsifiable in the short run, which is why the supply half of the conversation carries more information.

The most interesting strategic claim in this section is about the coding bet. Sacks is candid that he does not know why Anthropic went all-in on code, offering two competing explanations: an ideological bet that coding is the fastest path to recursive self-improvement, or a commercial read of Cursor's utilization data. Jason supplies the third possibility, that Anthropic watched Cursor's usage and decided to take the business. Baker reframes it as a general pattern, that the company moves into verticals where its own users are succeeding on top of the platform.

The competitive consequence is stated with unusual precision. OpenAI has reportedly pivoted toward coding, improved with GPT 5.6, and accelerated from 3x-4x annual growth to something over 20% month over month, which annualizes to roughly 10x. That is the strongest evidence in the episode that the frontier is contestable, and it cuts against the framing of an Anthropic monopoly. Sacks's conclusion is deliberately modest: Anthropic will grow fast regardless, and the open question is relative share rather than absolute growth.

Jason then narrows the question to infrastructure, asking what it would physically take to go from $100 billion to $1 trillion in a year. That question is where the episode stops being a valuation conversation and becomes an industrial one.